Steep-Slope CuInP₂S₆ Ferroionic Threshold Switching Field-Effect Transistor for Implementation of Artificial Spiking...

Area of Science:

• Materials Science
• Neuroscience
• Computer Engineering

Background:

• Spiking neural networks (SNNs) offer energy-efficient, biologically inspired computing but lack hardware development due to insufficient spiking neuron devices.
• Existing hardware for SNNs faces limitations in efficiency and mimicking neuronal dynamics.

Purpose of the Study:

• To present a novel CuInP2S6 (CIPS)-based threshold switching field-effect transistor (TS-FET) as a potential energy-efficient spiking neuron device.
• To demonstrate the device's capability to mimic essential neuronal dynamics and its application in SNNs.

Main Methods:

• Fabrication and characterization of a CIPS-based TS-FET.
• Evaluation of the device's electrical properties, including subthreshold swing and on/off current ratio.
• Integration of CIPS devices into an SNN for face classification tasks.

Main Results:

• The CIPS TS-FET exhibited ultra-steep switching (SS ≈7.5 mV dec⁻¹), high on/off ratio (>10⁷), and ultra-low off current (≈0.3 pA).
• The device successfully emulated leaky integrate-and-fire, threshold tuning, and spatiotemporal dynamics without auxiliary reset circuits.
• An SNN using CIPS devices achieved 95.83% accuracy in face classification via unsupervised learning and lateral inhibition.

Conclusions:

• CIPS TS-FETs show significant promise as energy-efficient spiking neuron devices.
• The ferroionic properties of CIPS enable efficient spiking and neuronal dynamics emulation.
• This work advances the development of next-generation SNN-based neuromorphic computing systems.